There are clearly numerous societal and environmental benefits to recycling municipal solid waste (MSW), including conserving natural resources, decreasing the amount of waste sent to landfills, and reduced environmental pollution.  If done efficiently, there can also be some worthwhile financial rewards. Although the recycling rate in the U.S. has increased massively since the 1960s, progress has stalled over the past decade1, and by many measures the process on a national scale has fallen short of its vast potential and magnanimous intentions.

Reasons for the stalled growth are complex and far-reaching

For one, waste by itself has negative value. Its disposal costs money. However, if single stream waste is separated into homogenous groups of paper, plastic, metal, etc., and only minimal contaminants remain in the various streams, what once was waste has value as commodities.  But it takes great effort to separate, so it must be done efficiently and cost-effectively.

Secondarily, Materials Recovery Facilities (MRFs) that process material for recycling have difficulty hiring and retaining employees to perform the dull, dirty, physically stressful and potentially hazardous work of manually separating waste.  Quite simply, there are a lot more appealing jobs available for the wage levels that make manual MSW sortation economically viable.

Automation and Artificial Intelligence make MSW recycling scalable and profitable

At a typical MRF (pronounced “murph”), mixed recyclables travel through a sophisticated maze of manual and mechanical sorting processes. Once the paper, cardboard, metals, and plastic are separated into like groups, they are compressed into large bales and sold to recycling companies to be converted into raw materials for new products.

National Recovery Technologies (NRT), based in Nashville, Tennessee, has developed a robotically automated system that handle a good portion of the waste sortation. Utilizing vision and Artificial Intelligence (AI), the NRT system improves productivity while removing people from the most hazardous, monotonous and undesirable areas of the process.

NRT was founded in 1981 by several Vanderbilt University PHD holders who had a mission to improve the environmental sustainability and circularity of waste management.  NRT was purchased in 2012 by Eugene, Oregon-based Bulk Handling Systems (BHS), an engineering and manufacturing company that has built some of the largest and most durable MRFs in the world.

A main point of differentiation between NRT and other waste sortation systems is NRT’s proprietary Max-AI® technology. Max-AI is a sophisticated neural network-based artificial intelligence (AI) platform that is used to identify objects in the waste stream.  The Max-AI vision system is used as a stand-alone detection device, integrated with robotics, or combined with NRT near infrared (NIR) optical sorting units to drive improvements in material recovery, operational efficiency, system optimization and maintenance.

Back in early 2017, NRT installed the first industrial robot integrated with the Max-AI technology, and now has more than 225 robotic waste recycling sortation installations across over 75 locations, on five continents.  All designed, engineered, manufactured, and tested in their Nashville location, supported by global service operations worldwide.

Burrtec Waste Industries – a major commitment to robotic sortation

In the spring of 2021 NRT installed 23 ABB IRB FlexPicker® delta robots equipped with Max-AI at a Burrtec Waste Industries MRF in Riverside, California. It is believed to be the largest installation of robots sorting waste in the world.

Burrtec is a privately-owned waste disposal company with several facilities of varying types serving the southern California market. The facilities are of varying ages with equipment of varying levels of sophistication and condition.

The 15-year-old Riverside MRF needed an upgrade and Burrtec approached BHS seeking a proposal on how to improve the Riverside facility, specifically asking about the latest technology. BHS personnel reviewed their facility to develop a plan to improve productivity, bringing in NRT to determine if the advanced Max-AI integrated robotics was viable for what Burrtec wanted to accomplish.

“We initially started working with BHS on the design and concept for the Riverside project, and they introduced us to NRT and their robots and Max-AI.  The technology seemed to have advanced enough to be worth considering,” said Richard Crockett, Burrtec Director of MRF and Transfer Operations.

“NRT did some testing by putting vision systems on our MRF lines to develop a baseline of the material we were processing, and to see if the AI-enabled robots could handle it.  The results were very positive, so we signed off on the Max-AI system and the 23 sortation robots as a primary technology component of the project. Prior to the installation of the robots all the waste sorting at the Riverside facility was performed manually.”

BHS also manufactured and commissioned many of the non-robotic components of the upgraded Riverside facility, including a significant portion of the new structures, conveyors and motors.  NRT’s portion was the technology package, the design, specification, installation and programming of the optical sorting, robotics, Max-AI vision systems and related system controls.

Overview of the Burrtec Riverside sorting operation

Collection trucks bring the MSW into to the Riverside MRF and the facility then works to separate everything out, utilizing a series of conveyors to route the waste through a multi-step sorting process, in the following order:

  • Waste initially passes through a series of disc screens, which first separates paper & containers from fines, then separates two dimensional material (e.g.  paper) from three dimensional material (e.g. plastic and aluminum containers).
  • Optical sorters then perform a secondary sort, utilizing air blasts to further separate material coming down the conveyors, routing material in one of two channels depending on material type.
  • The material then passes through stations comprised of a varying number of AI-enabled robotic sorters, which, depending on the nature of the stream, either continue cleaning valueless contamination from a material stream bound for recycling, or remove a valuable recyclable material from a valueless material stream bound for a landfill.

The optical sorters work at a very high throughput, and the robots conduct a final Quality Control sort at lower throughput, providing an important final cleanse to assure that the materials are completely separated into homogenous streams.

The recycle-grade material, once all sorted into defined groups, is bunkered and baled for shipment to market.

The robots and how material is processed

The Max-AI equipped FlexPicker robots are all labeled with an AQC prefix, which stands for “Autonomous Quality Control.” They are deployed in groups of one, two or four robots, which are noted with designations of AQC-1, AQC-2 and Dual ACQ-2, and are allocated based on how many robots are needed to handle the tonnage (volume) of a particular stream

“The role of the AQC robots is to do what a human does in terms of ensuring that material streams are as free of contaminants as possible, or any valuable commodities in a material stream bound for a landfill are recovered,” said Thomas Brooks, NRT Chief Technology Officer.

“The speed of the robot conveyors is set at 200 feet per minute, the conventional speed for manual sorters in the US.  If you have a lot of material coming down, you don’t slow down the belt, you add robots.  If there are 20, 30, 40 tons of material per hour you design the system to handle it within the parameters of the budget, physical space, etc.”

The 23 FlexPicker robots at the Riverside facility are deployed as follows:

  • Four Dual AQC-2s (16 robots) perform final quality control on a mixed paper stream spread across four conveyors, picking higher value brown fiber and grouping it with the cardboard, and picking out plastic containers and contaminants missed by the previous processes.
  • Three robot clusters sort contaminants from various plastic container streams. The material on these lines then goes beneath a magnet that pulls off any ferrous material, and an eddy current device that picks out any aluminum.    
  • An AQC-2 on a high-density polyethylene (HDPE) plastic container stream.
  • An AQC-2 on a general plastic container presort stream.
  • An AQC-1 on a polyethylene terephthalate (PET) container stream.
  • An AQC -1 on an aluminum stream.
  • An AQC-1 on a residue recovery stream that is bound for a landfill.  It takes one last pass at recovering any valuable recyclable material that had yet to be sorted.

Once a robot picks material from a stream, there is a series of chutes to the side of the conveyors that the robots drop the material through, and there is either a bunker underneath or oftentimes a conveyor to take the material away from the chutes.

If the robots are picking commodities from the mixed material stream, that valuable material is routed to be recycled with its appropriate material group.  If the robots are picking misplaced material from a recycling line for a specific material (e.g. picking textiles from a PET line), that material is also routed to be recycled with its appropriate material group. If picking contaminates from the material stream, that valueless material is typically routed to a holding area bound for landfill.

How it looks like:


The partnership with ABB and RobotStudio

As NRT was developing the Max-AI technology, they vetted a number of robot OEMs before deciding on ABB.

“We were looking for a robotic supplier that could give us three core pieces, and ABB checked all the necessary boxes,” said Brooks. “The most important is a global service reach, as we have installations on five continents. We are a small and lean operation, so the global support is critical, and ABB has a massive global footprint.

“A close second is engineering support and ingenuity on the front end, and a big part of that is ABB’s RobotStudio® design and programming software.  Diversity of product line is also important.  The Burrtec facility has all delta robots, but many of our other MRF installations utilize 6-axis articulated arm robots and collaborative robots.”

RobotStudio was a valuable tool in expediting NRT’s adoption of  AI-enabled robotic technology. It allows very realistic simulations to be performed, using real robot programs and configuration files identical to those used on the facility floor. With RobotStudio, Brooks and team are able to design the optimal AQC robot layout, and verify its performance before integrating the robots in a particular operation.

In order to achieve scalability, NRT’s approach with products, in general and with robot applications, is to build a standard product, and carry that forward to the waste stream. It may need some additional tailoring or adjustments when it gets into the field, but the core offering is standardized.

With RobotStudio NRT is able to able to simulate the design and performance of a complex robot installation before an end-user commits to the capital investment. By testing various design and set-up scenarios, NRT can determine the optimal robot system layout and determine how the robots can access and handle stream of widely non-uniform materials.

“We use RobotStudio to basically test all the different fringe cases and setups and everything else. Even to the point of driving pick rates and dealing with kinematics in order to get close to the belt, it all gets modelled in the development stage of the system, rather than at the deployment stage,” said Brooks.

“We leveraged RobotStudio to build an overlay application to simulate the variability of waste as it comes down the stream. We have used the overlay extensively to tie in the vision systems, and to map out the kinematic models of how to move waste from points A to B.”

The technology behind Max-AI   

NRT’s Max-AI Technology is combined with four Optical Sorters and the 23 AQC Robots at the Burrtec facility.  Max-AI is driven by a Visual Identification System (VIS) and a multi-layered neural network (NN) that sees and identifies more than 256 categories of waste material.

The neural network uses deep learning detection techniques, which identify classes of objects by a collection of complex interconnected features that are visually seen.  The neural network recognizes an object by millions of details that are inherent to a specific type of object.  This is extremely powerful as the neural network can infer differences in features to detect objects correctly.

The VIS camera is stationed above the conveyor, and ideally the material passing below on the conveyor is singulated to a degree that the camera can get a clear picture.  There are different NNs developed for different applications, so the NN used on the robots is not the same as used on the high-speed units connected to the optical sorters.

“AI is really the eyes and brains of the operation.  Just as a human sees something and uses its brain to take the appropriate action with its hands, the robots sees, thinks and acts in much the same way. The VIS is the seeing, with the support of the neural network, the AI is thinking, and the robot is taking the action,” said Brooks.

“On each of the six Burrtec material streams where robots are installed, Max-AI is making a decision, to either pick the material or not. If a material is to be picked, it could be a contaminate from otherwise viable and valuable streams of recyclable material; or it could be pulling out valuable recyclable items from a residue stream bound for a landfill.”

 Performance Improvements / Metrics

The Riverside facility renovation has driven significant efficiency for Burrtec, not only in labor savings and safety, but also in the recyclable commodity value they are extracting from the facility.

“With respect to the efficiency of this new facility in Riverside, we get the benefit of a higher throughput from the system, and an increased ability to process material. Prior to the upgrade, we were running our system anywhere from 20 to 22 tons per hour, and now we are doing 28 to 30 tons per hour,” said Crockett.

No employees were displaced as a result of the project, as Burrtec has always worked to “right size” their organization.  Employees who performed the manual waste sorting were moved to other, more appealing jobs within the organization.

“We now have a more efficient operation from a labor standpoint. The robots come to work every day and they’re not as susceptible to some of the things that could be found in the recyclables that could, injure or sicken a worker. So there’s some definite benefits with the robots, believe me,” said Trevor Scrogins, Burrtec VP of Operations.

Contributing to Environmental Sustainability

BHS and NRT have helped Burrtec and other MRF organizations, equipped with AI-enabled optical and robotic sorting, to recycle MSW in a scalable in a profitable way. From an environmental standpoint, diverting waste commodities from landfills and making them valuable, positively contributes to circularity and sustainability.

“People may challenge how effective recycling is, but nobody is against it. You read stories where only 10% of plastic is being recycled, which may be true, but the denominator keeps getting bigger. And with more automation I think that percentage will improve, and operations like ours will continue to make positive contributions to a more environmentally healthy society,” said Brooks.

Title: Source ABB